Collapsed Core vs A/D/C

[Jellyman_4eva]

Hi,

I am just wondering when it is best to use a collapsed core as opposed to the traditional three layers... Are there any rules of thumb on this? CISCO seem to only recommend collapsed core for a single building environment, however a lot of talk online seems to dictate that A/D/C is an expensive and theoretical best practice, and most people collapse...

As I have mentioned on a few posts I have a multi building site which I am looking to design for...

I am looking at using L2 access switches linked to two  L3 distribution switches per building... Now the question is...

Do I link these distribution switches to a pair of core switches in one"administrative" building (Which CISCO seems to suggest), or should I simply connect them all to each other...

The other thing I have noticed is that ring topologies do not seem to be in favour anymore?

I was thinking that the two real options were A + 2 * D in a building, with a central core (2 * core devices in the same room) (Creating a kind of star) (CISCO recommended)... or A + 2 * D/C in one building with D/C being one device (collapsed core), then per D/C device, take one link to the next building one way, and take the other link to the next building the other way, and once all hooked up creating a ring??

This seems to be a more resilient solution than two core devices lumped together in one building? But I have not seen this design recommended anywhere so was wondering if someone could point out why?!

[andrew.prince]

I generally find that the question aswers its self when you decide on what you want to achive redundancy/routing/failover etc.  Do you need a switch that just has users connected to also route?  If you have central WAN connections with simple IP Subnet deployments - do you need a distribution layer?

All hangs on what you are aming for - I personally do not think there is a right way or wrong way, providing you know what you are doing.

HTH>

[Jon Marshall]

It depends on what services for the campus you are running eg.

if you have servers where would you connect them. In a 3 tier architecture you would connect them to their own set of switches and then to the core. Without a core you could either -

1) connect them directly into a pair of the distribution switches

or

2) connect them to their own switches and then to a pair of distribution switches

either way you have now made a pair of distribution switches for a particular building not only responsible for routing the clients in that building but also responsible for campus wide services so those dsitribution switches would need to specced accordingly and bandwidth to that building would need to specced accordingly as well. The same argument could be made for internet connectivity/WAN connectivity etc.

You have in effect made these distribution switches a "pseudo core" but without the physical/logical separation. And you would need to ensure that this particular pair of switches had the highest levels of redundancy because they have become so important to the campus. This is also the issue with the ring setup. You generally want the shortest path from A -> B. With a distro/core it is generally 1 or 2 hops (depending on L2 or L3 core) from one distro to another. With a ring it would very probably be more.

The 3 tier model is designed so that you can logically separate services/functions within your network. Collapsing it blurs the edges.

Having said all that, i am not someone who believes you should slavishly follow design guides as long as you know what you are doing. The 3 tier model has been proven to work over many years but that doesn't mean you always have to deploy it in all situations.

Jon